How CarHunch Works

The Pipeline

You enter a registration — Type any UK vehicle registration. We accept cars, vans, motorcycles — anything with an MOT history.
We fetch the vehicle record — We query our database of 50M+ MOT records stored in high-performance columnar format (DuckDB + Parquet). Vehicle details, every MOT test, every defect, every advisory — retrieved in milliseconds from pre-built lookup tables.
We find comparable vehicles — Using the vehicle's make, model, fuel type, engine size, and manufacture year, we identify up to 200 comparable vehicles. For popular models (Ford Focus, VW Golf, etc.) there are typically hundreds of matches. For rare vehicles we widen the criteria progressively to ensure a meaningful comparison group.
We analyse defect patterns — We query pre-aggregated defect pattern tables covering the vehicle's peer group. This gives us the top 30 most common defects, their frequencies, severity distribution, and how they trend over time. The aggregation is done in advance so results are instant.
AI analyses the data — Our prediction engine uses sentence-transformer embeddings (MiniLM) to understand defect descriptions semantically — not just as keywords. "Brake pad worn below minimum" and "brake pad thickness inadequate" are recognised as the same underlying issue. This lets us detect patterns and predict likely future defects with confidence scores.
We score the risk — Four factors combine into a 0–100 risk score: how this vehicle's failure rate compares to its peers; how its defect frequency compares; whether it has any dangerous defect history; whether its mileage trajectory is normal. The result is a clear verdict: lower risk, typical, or higher risk — with evidence.
You get the full picture — Everything is presented in a single dashboard: vehicle details, MOT timeline, defect analysis, peer comparison, AI predictions, risk score, and an actionable summary. The whole analysis typically completes in 2–4 seconds.

Our Data

Source	Official DVLA MOT data — the same data that powers the government's MOT check service
Volume	50M+ individual MOT test records covering millions of UK vehicles
Freshness	Updated daily via an automated pipeline. New MOT results are typically available within 24 hours
Storage	Columnar Parquet format with DuckDB query engine — designed for fast analytical queries over large datasets
Derived tables	Pre-aggregated lookup tables for vehicle groups, defect patterns, and comparison statistics. This is why results are instant rather than taking minutes

Our Technology

DuckDB + Parquet	In-process analytical database reading columnar files. No network latency, no database server — queries run directly against the data files at analytical speed.
MiniLM embeddings	Sentence-transformer model that converts defect text into semantic vectors. Enables similarity matching that keyword search can't achieve — "tyre tread depth insufficient" and "tyre worn beyond legal limit" are understood as the same issue.
Redis caching	Results are cached so repeat lookups are instant. Comparable vehicle data, defect patterns, and full analyses are stored for fast retrieval.
Async worker pipeline	AI analysis runs in a dedicated background worker process. You see progress updates in real time as each stage completes — typically 2–4 seconds total.
Daily delta pipeline	Every morning, new DVLA data is downloaded, validated, converted, and merged into our tables automatically. No manual intervention, no stale data.
Pre-computed aggregations	Group statistics, defect patterns, and comparison baselines are pre-built into derived tables. When you search, we look up results — we don't recalculate from scratch.

What We're Not

1

We don't scrape

Our data comes from official DVLA sources, not scraped from other websites.

2

We're not opinion-based

No reviews from owners, no forum posts, no subjective ratings. Everything is derived from structured test data.

3

We don't sell data

Your searches are not shared with dealers, insurers, or third parties.

4

We don't charge for basic lookups

MOT history and vehicle details are free. AI analysis and comparisons are included.

5

We don't make absolute claims

Our predictions are statistical likelihoods, not guarantees. We always use comparative language ("higher risk than average") rather than absolute statements ("this car will fail").

Speed & Reliability

1

Vehicle lookup

< 1 second (pre-built derived tables)

2

Comparison data

< 1 second (pre-aggregated group statistics)

3

Full AI analysis

2–4 seconds (parallel pipeline with real-time progress)

4

Data freshness

Updated daily (automated delta pipeline)

Explore our Reviews, Comparisons, MOT History, and AI Hunches pages to see each feature in detail.